摘要
针对现有图像去雾方法采用单一大气光值求解、复原图像难以兼顾亮度与远景去雾效果的问题,提出一种融合全局与区域大气光值图的暗通道图像去雾方法.首先提出一种基于最小方差投影的大气光估计方法,减少大气光估计值受极值点的影响,提高大气光估计精度;其次,提出一种基于场景深度的区域大气光估计方法,对不同景深区域独立求解大气光估计,引入景深信息,兼顾改善近景亮度与远景去雾效果;同时,将两者融合,按照大气光值图对高亮区域透射率进行调整优化,既增加了位置相关信息,又提高了区域间的相关性,增强了复原图像亮度的均匀程度,改善了图像质量.实验结果表明,提出的算法与多种文献去雾算法相比,能够较好地平衡复原图像亮度与远景区域去雾效果,有效提高复原图像能见度,雾霾浓度评价指标(FADE)、平均梯度、信息熵及模糊系数等指标均有显著提升,复原图像更加清晰.
Aiming at the problem that the existing image dehazing methods are solved by single atmospheric light value,which leads to the difficulty of balancing the brightness and remote dehazing effect of the restored image,the dark channel image dehazing method based on global and regional atmospheric light value map is proposed.Firstly,a method of estimating atmospheric light based on the least variance projection is designed,which reduces the influence of the extreme point on the estimated atmospheric light value and improves the estimation accuracy of atmospheric light value.Secondly,a regional atmospheric light estimation method based on scene depth is proposed.The atmospheric light estimation is independently solved for regions with different scene depth,and the scene depth information is introduced to improve the near-range brightness and the long-range defogging effect.At the same time,the above two methods are combined to adjust and optimize the transmittance of the high brightness area according to the atmospheric light value map,which not only increases the location-related information,but also improves the correlation between regions,enhances the brightness uniformity of the restored image and improves the image quality.The experimental results show that the proposed algorithm can better balance the image brightness and haze removal from the perspective,effectively improve the visibility of the restored image.In addition,the indexes such as fog aware density evaluator(FADE),average gradient,information entropy and fuzzy coefficient have been significantly improved,and the recovered image is clearer.
作者
黄鹤
吴琨
宋京
王会峰
茹锋
郭璐
He Huang;Kun Wu;Jing Song;Huifeng Wang;Feng Ru;Lu Guo(School of Electronics and Control Engineering,Chang'an University,Xi'an,710064,China;Xi'an Key Laboratory of Intelligent Expressway Information Fusion and Control,Xi'an,710064,China;UAV National Engineering Research Center,Xi'an,710072,China)
出处
《南京大学学报（自然科学版）》
CAS
CSCD
北大核心
2021年第4期551-565,共15页
Journal of Nanjing University（Natural Science）
基金
国家重点研发计划(2018YFB1600600)
装备预研领域基金(61403120105)
陕西省重点研发计划项目(2021GY-285)
陕西省自然科学基础研究计划面上项目(2021JM-184)
长安大学中央高校基本科研业务费专项资金(300102329401)
西安市智慧高速公路信息融合与控制重点实验室(长安大学)开放基金(300102321502)。
关键词
图像去雾
暗通道理论
大气光值图
透射率优化
最小方差投影法
image dehazing
dark channel theory
atmospheric light map
transmissivity optimization
minimum variance projection method
